LIME (Local Interpretable Model-Agnostic Explanations)

• Publisher: marcotcr
• Status: active
• Version: 0.2.0.0
• Release Date: 2020-04-03
• Date Added: 2025-11-24
• Source URL: https://github.com/marcotcr/lime

---

Summary

LIME is an explainability framework that approximates complex model behavior locally around a specific prediction using simple, interpretable surrogate models. Within the AI security and assurance stack, LIME supports transparency, debugging, and validation of model decisions by providing clear, human-readable feature attributions for individual predictions. Typical use cases include auditing high-risk systems, validating fairness concerns, and ensuring traceability for regulated or safety-critical domains.

---

Key Takeaways

• Produces local, instance-level explanations using interpretable surrogate models
• Model-agnostic: works with any classifier or regressor
• Helps identify misleading, biased, or unstable model behaviors near decision boundaries
• Useful for debugging model predictions in safety-critical or regulated workflows
• Complements global interpretability tools like SHAP by offering fine-grained local views

---

Related Code

TBD

---

Additional Sources

• XAI Using Lime — geeksforgeeks overview
• arXiv original research paper

---

Tags

explainability, model-interpretability, XAI, auditability, risk-assessment

---

License

BSD-2-Clause